Integrating social, economic, and environmental risk into flood management of aging dam infrastructure by combining cost-benefit and multi-criteria decision analyses

Management planning for aging dam infrastructure is typically conducted through the lens of a traditional costbenefit analysis, in which flood characteristics are related to implementation costs while lacking endogenous consideration of environmental risks (i.e., pollutant dispersion, habitat disruption) and social impacts (i.e., vulnerability, community buy-in, hazard resiliency). To address this gap, we integrate cost-benefit ratios into a spatial multi-criterion decision analysis 10 framework that amalgamates a suite of social and environmental criteria with stakeholder-defined weights and inundation outputs from standard flood control modelling. We use this framework to assess the costs and trade-offs for eight (8) alternative mitigation strategies associated with the Addicks and Barker Reservoir System in Houston, Texas, USA under extreme rainfall conditions. This case study illustrates how the total effectiveness of flood management scenarios may shift when flood modelling outputs are combined with spatially distributed environmental and social risks. We merge quantitative and 15 qualitative data for high-risk decision-making, thereby fostering stakeholder collaboration amongst conflicting goals.


Introduction
In the United States alone, there are over 90,000 artificial dams, including various flood control reservoirs, recreational lakes, water supply resources, and hydropower facilities, many of which were constructed with earthen materials following the U.S.
Flood Control Act of 1936. Urbanization and climate change has amplified the water pressure within these aging structures, 20 thereby increasing the risks for significant dam failure. Over one-third of the nation's dams have been classified as 'Significant-Hazard Potential', 'High-Hazard Potential', or completely 'Deficient', according to the level of structural integrity and the severity of consequences in the event of a breach (ASCE, 2017); thus, the impaired infrastructure systems must be strategically managed to reduce the risk of widespread flooding.
The severity of this issue was highlighted during Hurricane Harvey when two high-hazard flood control reservoirs, namely, 25 the Addicks and Barker Reservoir System (ABRS) in Houston, Texas, USA, were challenged by unprecedented amounts of rainfall. Typically, flood control reservoirs mitigate risk by storing large volumes of stormwater and systematically releasing flows through timed operations to minimize downstream impacts. However, the storage load during Hurricane Harvey posed a risk of catastrophic failure for the ABRS, which could have resulted in a wall of water damaging much of the metropolitan https://doi.org/10.5194/nhess-2021-144 Preprint. Discussion started: 10 June 2021 c Author(s) 2021. CC BY 4.0 License.

Integrated flood management
Traditional flood management involves the narrowing-down of numerous mitigation options into a select focused array through 65 high-level flood modeling and a standard cost-benefit analysis (CBA) (Brouwer and van Ek, 2004). Social and environmental considerations are loosely considered in such approaches but are not included as dependent variables in the decision-making framework. Instead, a preferred mitigation option is often chosen from the primary perspective of flood inundation extents and economic costs. Detailed site assessments are then performed during the engineering project phase. For example, in the latest planning study published by the USACE for the ABRS system, economic costs and the extents of flood protection formed the 70 fundamental basis for project recommendations. Environmental resources were introduced qualitatively, with a description of potential impacts to natural habitats. Environmental service facilities (i.e., treatment and industrial facilities) were mentioned only briefly and were largely indeterminate. Socio-demographic factors were also summarized at a high-level for each watershed boundary (population, income, education, and race) (USACE, 2020). However, a quantitative method for including such considerations into the overall decision-making process and how such risks could influence the proposed alternatives was 75 lacking.
Flood reservoir planning may benefit from the use of a multi-criterion decision analysis (MCDA) approach for evaluating alternatives through explicit consideration of societal and environmental risks. MCDA is a blanket term used to describe a variety of methods that evaluate multiple, and often overlapping, criteria in a structured decision framework (Voogd, 1982).
Common MCDA approaches within the hydrological literature include analytic hierarchy processing (Le Cozannet et al., 80 2013), fuzzy set theory (Li, 2013), multi-attribute utility theory (Yang et al., 2012), simple additive weighting (Liu et al., 2014;Sjöstrand et al., 2018), evolutionary optimization (Karterakis et al., 2007), and TOPSIS (Lee et al., 2013) (Velasquez and Hester, 2013). Researchers are increasingly trending toward participatory modeling for flood vulnerability assessments and management approaches due to the numerous qualitative benefits gained through stakeholder involvement (de Brito et al., 2018;Karjalainen et al., 2013;Ronco et al., 2015). Here, we focus on the application of spatial MCDA by employing a simple 85 additive weighting (SAW) approach using stakeholder feedback in a geographic information system (GIS) environment. Spatial MCDAs are used to integrate overlapping criteria with stakeholder values, which have gained popularity with the increased availability of geographic datasets (Malczewski and Jankowski, 2020). GIS-based MCDA tools allow consideration of the tradeoffs involved in different scenarios and visualization of how decisions impact the region.
In the context of flood risk management, spatial MCDAs have been largely used to evaluate the overall net impact of flood 90 magnitude for alternative mitigation measures. The interdisciplinary application of MCDAs for social vulnerabilities and environmental risk assessment has received considerably less attention (de Brito and Evers, 2016;Malczewski, 2006). MCDAs have become increasingly popular in water resources planning and management (de Brito and Evers, 2016;Hajkowicz and Collins, 2007), but such approaches for flood-control dams have primarily focused on optimization of release operations and not the planning of new structures (i.e., Chu et al., 2015;Fu, 2008;Labadie, 2004;Zamarrón-Mieza et al., 2017). The extension 95 of MCDA approaches to consider large-scale infrastructure planning for dam systems is a largely undeveloped area of research (de Brito and Evers, 2016;Zamarrón-Mieza, 2017). As we experience a continued shift toward more extreme storm events and aging dam infrastructure, we necessitate a planning framework that evaluates risk comprehensively. Figure 1 describes the traditional decision-making workflow for flood management, as exemplified in the latest USACE planning study for the ABRS (USACE, 2020), while contrasting our proposed workflow that combines spatial MCDA techniques with CBA and 100 inundation modeling.

Case Study: The Addicks and Barker Reservoir System 105
The ABRS comprises several watersheds in the Houston region that are hydrologically-connected via the Addicks and Barker flood management dams and their downstream releases into Buffalo Bayou, as well as cross-basin overflow from Cypress Creek that enters the reservoir watersheds during extreme events (Fig. A2). The reservoirs are operated by the USACE and have been classified as two of the most-hazardous and deficient dams in the United States (USACE, 2010). During Hurricane Harvey, large volumes of water spilled over the Cypress-Addicks watershed divide and entered the local reservoirs, introducing 110 significant uncertainty regarding the ability of the reservoirs to withstand the stored pressure (Sebastian et al., 2017). To accommodate these increased inflows, and to reduce the risk of catastrophic dam failure during Hurricane Harvey, the https://doi.org/10.5194/nhess-2021-144 Preprint. Discussion started: 10 June 2021 c Author(s) 2021. CC BY 4.0 License. reservoirs were released according to unprecedented surcharge procedures, causing widespread flooding in the receiving channel and damaging thousands of structures (HCFCD, 2020;USACE, 2017). Simultaneously, overland flow conditions in the adjacent watersheds interacted with the reservoir releases and compounded regional flood conditions. Flooding associated 115 with Hurricane Harvey damaged over 154,000 homes in the greater-Houston region, of which at least 46,800 flooded structures were located within the ABRS inner-connected watershed system (HCFCD, 2018). The extent of damages inspired widespread discussions regarding regional drainage infrastructure, with specific attention to mitigation of the ABRS reservoirs (USACE, 2020).

Proposed mitigation strategies 120
Following Hurricane Harvey, discussions regarding how best to address the flood risks associated with the ABRS were raised throughout the community. Popular mitigation solutions included dredging a large underground tunnel, adding an additional reservoir to capture cross-basin overflow, widening receiving channels, increasing storage capacity, optimizing release operations, and buying-out properties. Such strategies are reminiscent of the original 1940 ABRS project plan, where additional open space, reservoir storage, and routing improvements provided an added layer of protection but were later abandoned due 125 to limitations in funding and land availability (Fig. A2;USACE, 1940). In 2018, the Harris County Flood Control District (HCFCD) and the USACE commissioned a joint study of the ABRS system to evaluate various recommended solutions. An interim report was released in October 2020 describing preliminary recommendations for the ABRS strategies. Eight alternative scenarios were screened on the basis of CBA and narrowed to a focused array of five mitigation strategies recommended for further analysis (USACE, 2020, Table 3). We analyzed these eight preliminary alternatives through our 130 proposed CBA+MCDA framework to investigate the impact of considering comprehensive environmental and social risks in the screening phase, summarized in Table 1. Hydrological assumptions for each of the eight alternatives are described in Section 3.1. Table 1: Array of alternative mitigation strategies and cost estimates, in 2020$ million (M), from the October 2020 U.S. Army Corps of Engineers interim feasibility report (USACE, 2020). * : The diversion levee for cross-basin overflow was excluded from the USACE report.

135
Instead, we used the monetary value for diversions between adjacent watersheds in the USACE (2020) report (per linear unit) and extrapolated to assumed lengths for this case study. Underground tunnels to re-route existing reservoir outflows. 9,250 M

Environmental and social impacts
During Hurricane Harvey, highly industrialized regions of Houston were impacted and released various pollutants into the environment. The floodwaters distributed over one-million gallons of hazardous materials throughout the region (Christine 140 and Yue Xie, 2018; Miller and Craft, 2018), resulting in widescale and long-term health impacts from exposure to bacteria, chemical toxins, mold, and carcinogens (Horney et al., 2018;Kapoor et al., 2018;Schwartz et al., 2018;Stone et al., 2019).
Various environmental consequences were associated with ABRS releases, such as flooded wastewater treatment plants, leaking storage tanks, contaminant dispersal, sediment redeposition, and disturbance of acidic soils (Folabi, 2018;Kighadi and Rifai, 2019). Research following Hurricane Harvey also highlighted the varied social factors that contributed to disparate flood 145 impacts and resiliency among population groups. The Houston region lacks formal zoning and is comprised of a highlyheterogeneous composition of vulnerable populations interspersed with wealthier communities (Christine and Yue Xie, 2018).
Vulnerable populations are at higher risk of experiencing post-traumatic stress disorder and long-term health effects after a flood disaster and tend to be the slowest to recover, leading to endemic poverty issues (Dickerson, 2017;Grineski et al., 2020).
Wealthy and middle-income populations face higher risks when located outside of federally-designated floodplains where 150 flood insurance is voluntary (Dickerson, 2017). Mobility issues associated with flooding reduced access to emergency services, which posed additional hazards to vulnerable populations and led to several fatalities during Hurricane Harvey (Bodenreider et al, 2019, Chakraborty et al., 2019, Jonkman et al., 2018. In considering the proposed ABRS mitigation strategy of adding a third reservoir, local concerns have been raised regarding the environmental disruption of conserved prairie lands that provide natural stormwater mitigation, recreational opportunities 155 (TPL, 2018), and habitat preservation, including the federally-endangered prairie dawn-flower (FWS, 2021). The proposed strategy of channelizing the ABRS receiving stream also presents habitat disruption for a highly-threatened species, the Alligator snapping turtle (Munscher et al., 2020), and other aquatic biota. We noted negative social connotations associated with channelizing the natural stream, which has long-been opposed by local residents due to recreational and amenity preferences (Schafer, 2013). We also considered a high societal risk for the mitigation option of diverting cross-basin overflow 160 to an adjacent watershed, which has been shown to increase community flood risk downstream (GHFMC, 2019). While these issues have been studied as individual occurrences, there exists a limited understanding of how such factors interact holistically and impact cross-basin mitigation planning. As such, the practical integration of environmental and social systems into flood risk management requires further efforts for improved resiliency.  (Table S1) and the Green and Ampt method in the Buffalo Bayou watershed (Table S2). Peak flows from the HEC-HMS output hydrographs were used as inputs to the HEC-RAS models (Table  180   S3 --A1 (Baseline): Models were downloaded from the HCFCD M3 system, updated using latest geospatial datasets, and calibrated to field observations for comparison to alternative mitigation strategies.
--A2 (Additional Reservoir): We assumed the maximum storage volume of an additional reservoir at the Cypress-Addicks drainage divide was 2.34x10 8 m 3 (190,000 acre-foot) with a 56.6 m 3 /s (2,000 ft 3 /s) outflow near the Bear 190 Creek tributary, per (HCFCD, 2015 andUSACE, 2020); however, our model only captured approximately 1.23x10 8 m 3 of flow (100,000 acre-foot) under Hurricane Harvey conditions (see Appendix Text A2). We linked the Cypress Creek hydrological model with the Addicks model by simulating diversion nodes to capture the estimated quantity of cross-basin overflow (Fig. S5).
--A3 (Addicks Watershed Buyouts): In this scenario, we assumed approximately 10,000 homes located below the 195 Addicks Reservoir spillway are purchased and reallocated. We therefore adjusted the curve number values in each of the impacted subbasins to account for an increase in open space (Table S1).
--A4 (Diversion Levee): We assumed that one-half of the transfer observed during Hurricane Harvey continued as overflow into the Addicks watershed, while the other one-half of overflow was diverted to Cypress Creek by a levee. We incorporated assumptions regarding the risk of increased flooding in the lower portions of Cypress Creek by 200 adopting inundation bounds from GHFMC (2019), which modeled a diversion levee at the Addicks-Cypress Creek watershed divide.
--A5 (Buffalo Bayou Buyouts): We assumed that 441 structures along Buffalo Bayou were acquired in this scenario (per USACE, 2020) and modified the associated subbasin parameters to account for these changes. We removed associated land parcels from the composite social risk maps in this alternative. 205 --A6 (Increased Storage): Here we assumed that an increased storage capacity within the existing bounds of Addicks and Barker Reservoirs would allow for an optimized release strategy into the receiving channel. We adapted release rates from the USACE 2010 Interim Control Action Plan (USACE, 2010), where constant releases began once the stream flow at the Piney Point gauge reached 113.27 cms (4,000 cfs) (see Appendix Text A2).
--A7 (Enlarged Receiving Channel): We modified the Buffalo Bayou channel geometry within the HEC-RAS model 210 to capture an additional 340 m 3 /s (12,000 ft 3 /s) capacity. In this scenario, the modified channel cut depth averaged 3.0-4.6 m (10-15 ft) with 1V:4H side slopes, daylighting to natural ground. The proposed top widths in our channel geometry ranged from approximately 100-160 m (350-540 ft).
--A8 (Underground Tunnel): Here we assumed no outflows from the Addicks or Barker reservoirs into the receiving channel, thereby removing inflows from the HEC-HMS source nodes for Buffalo Bayou. Per USACE (2020), 215 reservoir storage water would be re-routed around the city and toward Galveston Bay in this mitigation scenario.

Spatial multi-criteria decision analysis
Here, a GIS-based MCDA simple additive weighting (SAW) approach (known as the "weighted overlay" method) was used to spatially link the hydrological and hydraulic modeling with environmental and social flood risk. SAW MCDA approaches establish value functions by multiplying each criterion by a specific weight and summing the total numerical scores. SAWs 220 have the advantage of being intuitive for decision-makers and have observed wide application throughout the hydrological literature (de Brito and Evers, 2016;Velasquez and Hester, 2013). Typical weighted overlay steps include: 1) Define the problem, goal, or objective holistically. 2) Determine criteria and constraints from local sources and expert opinions. 3) Standardize factors into a common scale through reclassification. 225 4) Rate and weight the importance of each factor. 5) Aggregate layers and criteria into an overall suitability map. 6) Apply constraints, as applicable. In Step 1, the study objective was defined as identifying the mitigation alternative that minimizes economic, social, and 230 environmental risks while providing maximum flood protection benefits. The contributing criteria (Step 2) were determined from a literature search of environmental and social vulnerabilities exacerbated by flooding (Section 2.2), complemented with local knowledge regarding several specific alternatives. For Step 3, each of the environmental and social factors were converted to a standardized gridded dataset with a uniform scale from 0 to 100 (where higher numerical values represented greater risk).
To standardize the point and polyline feature classes into spatially varied datasets, the Euclidean Distance method was applied. 235 Euclidean distances convert feature layers into gridded datasets by assigning a value to each cell that indicates the distance of that cell to the nearest criterion, thus standardizing space and creating hotspots in multi-criteria decision making. In Step 4, the influence of each criterion was incorporated into the analysis by multiplying the standardized datasets by the weights in Table   2. The weighted layers were aggregated (Step 5) to produce composite risk maps within the ABRS system in terms of environmental and societal criterions and classified into levels from low to high risk. A flood inundation mask was then applied 240 to each composite index to constrain the functions to areas that would be impacted in the modeled scenarios (Step 6). Thus, the total risks analyzed for each alternative included a hybrid combination of environmental, hydrological, and societal factors.
Composite risk maps were derived by multiplying stakeholder-chosen weights (w) by normalized evaluation scores (e) for each identified criterion (j) using the weighted overlay method, Eq. (1): Where RE|S refers to the composite risk value of the gridded cells for each spatial map within the domain (E: environmental, S: social), n represents the number of criteria in each domain, wj refers to the relative weight of each criterion (j), and ej represents the normalized evaluation score (0 to 100).
Composite impact functions were obtained from Eq. (2) for environmental degradation and social effects using zonal statistics for the composite risk and the modeled inundation area of each alternative (i): 250 where IFE|S represents the impact function for each domain, 9 3|5 describes the average risk value in each zone, and ai refers to the zonal area impacted in each alternative. We defined ancillary risk as an additional measure of risk value specific to select alternatives, further described in Section 3.2.2.

Comprehensive risk datasets 255
An exploratory geospatial review was conducted to evaluate the applicability of various criterions identified in the literature search, including water quality point-sources, petroleum storage tanks, superfund sites, wastewater treatment plants, petroleum terminals, natural gas facilities, landfills, oil spills, air quality stations, mold growth, medical facilities, population growth, educational centers, emergency response locations, ethnicity, income, transportation, utilities, and damaged structures. These datasets were consolidated into the primary factors that spatially correlated to changes in flood patterns within the ABRS 260 system, shown in Table 2. Toxic release inventories, leaking petroleum storage tanks, wastewater treatment plants, and the potential for soil erodibility were used to create a composite geospatial map of environmentally-sensitive areas in the reservoir watershed system. Stream samples were obtained from field campaigns following Hurricane Harvey (Kiaghadi and Rifai, 2019), which were used in this study to validate the areas of environmental burdens associated with contamination in local waterways. The Center for Disease Control (CDC) maintains a Social Vulnerability Index (SoVI) that combines various risk 265 factors related to natural disaster preparation and recovery. The CDC SoVI has been used in various studies as a suitable index for flood hazard mapping due to the underlying criteria, including socioeconomics, household composition, disabilities, minority status, language, housing, and transportation (Flanagan, 2011). An independent analysis validated the CDC SoVI for use in disaster planning and adaptation (Bakkensen et al., 2017). For purposes of this study, the SoVI index was used as the primary factor in societal risk related to flood hazards. Additional social factors used in the MCDA included population density, 270 flood insurance, roadway inundation, and proximity to medical facilities. The spatial risk associated with flood insurance was derived from national flood hazard zones and a repository of damaged structures in the community. It was assumed that residents within the FEMA 1% and 0.1% flood zones carried flood insurance, while 20% of all other residents had purchased voluntary insurance (Klotzbach et al., 2018). The depth of roadway inundation was chosen as a limiting mobility factor for access to and from emergency services. Water inundation was exported from pre-defined hydraulic modeling ensembles and 275 used to select roadways that would be inaccessible.

Ancillary risk datasets
Several mitigation alternatives included ancillary risks for land preservation, biodiversity, and social impacts. Location assumptions for the additional reservoir and cross-basin overflow patterns were georeferenced from HCFCD (2015) to estimate the spatial flood difference. For Alternatives A2 and A7, the areas of impacted conservancy lands (prairies and 290 channel easements, Fig. 3) were assigned 9 3 = 100 and integrated into the weighted IFE calculations. Similarly, the composite social risk IFS for A7 included 9 5 = 50 for the natural recreational areas disturbed by channelization. In Alternative A4, the social impact associated with increased flooding along Cypress Creek was assigned 9 5 = 100 to quantify the added communal stressors from worsened downstream conditions. Finally, we included 9 5 = 50 for the areas of property buyouts along Buffalo Bayou in Alternative A5, to quantify how this region is comprised of established neighborhoods who have strongly resisted 295 buyout efforts in the past (Houston Strong, 2020).

Weight determination 300
We assigned general weights for each criterion (Table 2) according to discussions with Houston-area flood risk stakeholders, including governmental entities, interest groups, and specialized consulting firms. This knowledge-based approach follows the participatory planning nature commonly associated with MCDA SAW frameworks. In participatory decision-making, external stakeholders collaborate to guide the impact and importance of the multiple criteria used in the MCDA (Mendoza and Martins, 2006). Structured approaches for quantifying the evaluation criteria from stakeholder judgements include the Delphi technique 305 (Lee et al., 2013;Pathak et al., 2020), the step-wise assessment ratio (Khosrayi et al., 2018), focus groups, and cognitive mapping (Mendoza and Martins, 2006). As participatory modelling is inherently qualitative, individual criterion weights will differ according to local conditions and stakeholder goals. We presented the weights in Table 2 as a proof-of-concept for the CBA+MCDA framework; we, therefore, anticipate formal adoption of this framework might explore several of the additional participatory modeling methods available per local resources and goals. By considering multiple, oftentimes conflicting sets 310 of criteria, participatory weighting coalesces quantitative techniques with qualitative expert opinions to facilitate discussions, elucidate internal values, and aid in justifying final decisions (Mendoza and Martins, 2006).

Cost-benefit analysis
CBA frameworks are used to compare multiple water policy options on the basis of economic efficiency, which is measured as the difference between added costs and benefits (Ward, 2012). In the context of classic flood management decision-making, 315 added costs include the implementation costs of any new mitigation strategies, and added benefits include a reduction in flood risk when compared to the status quo condition. Additional considerations often include long-term operational and maintenance costs, investment costs, economic growth potential, and non-monetary societal costs (i.e., community willingness, business disruption, housing relocation) (Jonkman et al., 2004). Cost-benefit indices were derived according to Eq. (3) using the estimated implementation costs for each of the mitigation alternatives (USACE, 2020) and the avoided flood risk area, 320 obtained from hydrologic and hydraulic modeling. Areas were converted to monetary units by averaging the cost per unit area in the traditional CBA calculations (USACE, 2020, Table 7), resulting in a flood inundation conversion factor of 0.478 $M/hectare for this case study.
where CBi refers to the cost-benefit indicator, Ci represents the implementation cost in 2020$ million, and Bi represents avoided 325 inundated area (inundated area A1 -inundated area Ai), converted to monetary units.

Integrated CBA+MCDA
We merged the CBA and MCDA approaches by ranking Alternatives A1-A8 for each set of indicators (CB, IFE, IFS) and assessing the results relative to each reference alternative. Since the unique indicators contained different units of measurement ($/hectares, 0-100 risk) we used z-score normalization to transform the values to equivalent scales. We aggregated the goals by assigning an equal weight to each of the three dimensions in our framework (economic, 335 environmental, social) and used a linear additive model to integrate the sub-indicators. We assumed mutual, preferential independence between each of the unique sub-indicators for the linear additive aggregation method; however, we understand this approach may not be useful in all mitigation scenarios, and we recommend the analysis team to consider outranking methods and geometric aggregation, as applicable (Nardo et al., 2005). We normalized each set of data using the z-score methodology and defined the total risk score according to Eq. (4) and Eq. (5) and 7 = KL7 KL + 37 3 + 57 5 , where Z(CB|E|S)i describes the z-score of the indicator for domain (CB: cost-benefit, E: environmental, S: social), w(CB|E|S) refers to the weighting of each domain (equal weighting of 1/3 in this case study), and Ti refers to total risk. 345

Hydrologic and hydraulic modelling
Hydrologic and hydraulic models were used to obtain flood inundation extents for each of the proposed ABRS mitigation alternatives, A1-A8. Alternatives A1 through A4 corresponded to changes in flood inundation primarily within the Addicks watershed (Fig. 4), and Alternatives A1, A5-A8 represented mitigation options that would impact the Buffalo Bayou watershed 350 (Fig. 5). Total simulated inundated areas for Alternatives A1-A8 (hectares) were: 6,774.92 (A1Addicks), 3,184.14 (A2), 6,308.72  (Table A2), with an R 2 coefficient of 0.99 and a Nash-Sutcliffe efficiency (NSE) of 0.95. The resulting flood inundation bounds also compared well 360 against manual spot inspection of flooded areas, per NOAA aerial imagery (NOAA, 2017b). These results demonstrate the accuracy of the model in capturing overland flow conditions within the watershed and the observed reservoir releases from Addicks and Barker.
--A2 (Additional Reservoir): As described in Appendix Text A3, we estimated approximately 1.23E 8 m 3 (100,000 acrefeet) overflowed from the Cypress Creek watershed into the Addicks watershed during Hurricane Harvey, which 365 corresponded well to observed flooding conditions. An additional reservoir would therefore need to capture at least this amount to minimize cross-basin transfer. The USACE (2020) resiliency study proposed an additional reservoir capacity of 2.34E 8 m 3 (190,000 acre-feet). Since we assumed a constant outflow from the reservoir into Bear Creek (HCFCD, 2015), the majority of flood inundation improvements between A1 and A2 were observed near this confluence. While the addition of a third reservoir improved conditions, a significant portion of the flows into the 370 Addicks Reservoir resulted from overland conditions within the watershed. By comparing outflow hydrograph volumes from HEC-HMS (Fig. A4), we estimated only 16.42% of the flows into Addicks Reservoir were accounted by the cross-basin transfer, thereby suggesting additional mitigation measures are warranted to limit substantial inflows to the aging dams.
--A3 (Buyouts): When considering the hydrological impacts for buyouts by altering the subbasin loss parameters, we 375 observed negligible changes to the overall peak flow conditions, primarily due to the location of the buyout parcels near the downstream bounds of the model. Since peak flows were used to drive the HEC-RAS models, the inundated areas for A3 were nearly identical to baseline conditions, suggesting an unreasonable improvement to flood conditions given the high estimated cost ($5B).
--A4 (Diversion Levee): While the USACE (2020) mitigation report excluded diversions from the focused array due to 380 the adverse flood impacts in adjacent watersheds, we chose to simulate model conditions for purposes of comparison in the CBA+MCDA framework. We demonstrated a significant reduction in flood inundation area within the Addicks watershed in comparison to baseline conditions (6,774.92 to 4,734.85 hectares). However, these benefits were partially by additional flooded area in the downstream portions of Cypress Creek. These findings highlight the hydrologically-interconnected nature of cross-basin dams and stress the necessity to consider social impacts as a 385 function of regional space.
--A5 (Buyouts): Modelling results for Alternative A5 were similar to observations for buyout conditions in the Addicks watershed. Specifically, the peak flow outputs differed a negligible amount when altering the subbasin loss parameters due to the location of the parcels near the receiving stream. We therefore removed the area of buyout parcels (in both A3 and A5) from the inundated area to represent limited flood impacts in these regions. 390 --A6 (Increased Storage Capacity): By increasing the storage capacity in the existing Addicks and Barker dams, we assumed optimized timing of reservoir releases into the receiving channel. Alternative A6 resulted in a significant reduction of flood inundation area by approximately 64% from baseline conditions (1,456.45 to 519.58 hectares). We noted the USACE (2020) mitigation study did not inherently link the provision of optimized releases by increasing storage capacity. Our findings suggest that additional capacity in the existing reservoirs may have alleviated the need 395 to release surcharged flows into the receiving channel during Hurricane Harvey by allowing stored floodwaters to remain within the reservoirs for a longer period of time (Fig. A3).
--A7 (Enlarged Receiving Channel): In order to achieve adequate storage of the observed releases during Hurricane Harvey, we needed to expand the top width geometry for Buffalo Bayou in our HEC-RAS model to extents much wider than what was proposed by the USACE (~100-160 m in our model vs. 70 m in USACE, 2020). While our study 400 is not intended for detailed design, significant displacement of land would be necessary for this alternative scenario, impacting social and environmental considerations along the banks of this natural stream.
--A8 (Underground Tunnel): We noted similar improvements to the inundated area for Alternative A8 (560.61 hectares) when compared with A6 (519.58 hectares), highlighting how the receiving channel is driven primarily by outflows from the dams. The significant costs associated with the tunnels ($6.5-12B) and the comparable flood mitigation 405 benefits to other alternatives led the USACE to drop this option from the focused recommendations (USACE, 2020); however, as described in Section 4.2, when we consider the added risks and benefits associated with social and environmental impacts, the tunnel alternative warrants further consideration.

Integrated risk analysis
By using the traditional CBA framework for the dam mitigation options, Alternatives A2 (Additional Reservoir) and A7 415 (Channel Improvements) resulted in the lowest CBi ratios for the Addicks and Buffalo Bayou watersheds, respectively. As such, the USACE (2020) interim study prioritized these strategies in a recommended focused array of optimal plans (A4 was not considered in the interim study). Alternative A5 (Buffalo Buyouts) was included in the proposed focused array only as a means for constructing the widened receiving channel (USACE, 2020). Shortly after publication of the interim dam study, a report was compiled by a local coalition of flood resiliency stakeholders highlighting the need for further attention to ecological 420 and social factors associated with the ABRS mitigation alternatives. These constituents urged the USACE to further explore Alternatives A6 (Increased Storage) and A8 (Underground Tunnels) in light of holistic environmental and social considerations. This report also stressed the negative social impacts associated with parcel buyouts for the socio-demographics represented along Buffalo Bayou, particularly if the buyouts are conducted in a patchwork manner (Houston Stronger, 2020).
In echoing these concerns, our study demonstrates how quantitative inclusion of social and environmental criteria can alter the 425 ranking of potential flood mitigation strategies. Table 3 summarizes the results of the case study calculations, Fig. 5 shows changes to the rankings between the CBA and the CBA+MCDA frameworks, and Fig. 6 compares these integrated outputs across the financial, social, and environmental domains. In comparing the CBi ratios for each of the alternatives, we see that the buyout alternatives (A3 and A5) provided the lowest benefit when viewed strictly in terms of flood benefits due to the hydrological properties of the watersheds, which resulted 435 in a minimal reduction of inundation risk (Section 4.1).
However, when we integrated environmental and social considerations among the alternatives, we noted a potential for tradeoffs in cumulative risk (i.e., A5 surpassed the rank of A7 due to the high environmental and social consequences 440 associated with channelization). In the Addicks watershed, our composite risk framework maintained the disinclination toward A3 by incorporating high costs of buyouts with a minimal improvement in social or environmental conditions. We noted a reduction in the relative preference of A2 445 (Additional Reservoir) when compared with A4 (Diversion Levee) for mitigating cross-basin overflow, with the Ti for each alternative nearly identical. If we were stakeholders during the planning of this case study, such a finding would encourage us to consider coupled flood modeling for the Cypress Watershed prior to discounting A4. By investigating the resulting composite risk maps for A2, which was a preferred mitigation strategy in the USACE (2020) report, we noted the flood risk was diverted largely to unpopulated, low-vulnerability areas with higher 450 likelihood for soil erosion potential, thus lowering the choice suitability. Moreover, as discussed in Section 4.1, the addition of a third reservoir here does not fully mitigate the flood issues with the Addicks watershed, which is driven largely by overland flow. By collectively viewing the synergies between these domains, we identified a need to further investigate alternative strategies for mitigating the cross-basin overflow.
In the Buffalo Bayou watershed, the preferred mitigation option in the CBA framework (A7 -Channel Improvements) 455 transitioned to the highest-risk alternative when considering CBA+MCDA. Alternative A7 would require substantial changes to Buffalo Bayou, which is highly opposed by residents, without necessarily providing additional improvements in comparison to other mitigation options presented here. Alternative A6 (Increased Storage) resulted in the most preferable option due to less relative ancillary risks than A7 and substantial flood reduction. Moreover, Alternative A6 provided opportunity for optimized timing of reservoir releases, an inherent benefit that was not considered in the USACE (2020) approach. The next 460 highest-ranked scenario in the Buffalo Bayou cohort was Alternative A8 (Underground Tunnels), which was initially discarded in the CBA framework due to the significantly high costs associated with subsurface construction. When incorporating the reduced environmental and social impacts conferred by re-routing flood water away from the urban centre, Alternative A8 became a higher-performing option, warranting inclusion in the focused array.  Due to the compensatory nature of the linear aggregation methods employed in this study, the comprehensive risks and subsequent rankings of alternatives will be sensitive to changes in the stakeholder-defined weights (i.e., Table 2, Eq. 5). We 470 suggest that a high sensitivity of weightings is advantageous at the screening stage of flood management, particularly when conflicting goals are present across diverse domains. This method of aggregation reflects the multiplicity of stakeholders' viewpoints and fosters an iterative and interactive approach to scenario-based management. By coupling the framework results directly with stakeholder values, we facilitate the need for greater cooperation among various interest groups. For example, following the ABRS interim study (USACE, 2020), several community groups had expressed concern regarding the CBA-475 based recommendations due to conflicting goals stemming from unique perspectives and expectations for the mitigation alternatives (Houston Stronger, 2020). When we integrated environmental and social factors into our CBA+MCDA framework, we noted a change in alternative rankings, which could be used to foster further discussion between the interest groups. This benefit to participatory modelling facilitates transparency regarding goal magnitudes, which helps decisionmakers gain a greater understanding of their own assumptions and values (Mendoza and Martins, 2006). The acceptability of 480 trade-offs between diverse criteria becomes easier to identify and justify, and stakeholders are provided a clear opportunity for active engagement in the decision-making process. The CBA+MCDA framework can then be performed in an iterative fashion by altering stakeholder weights and assessing outcomes until the communal goals of the stakeholder cohort is satisfactory (ref. Fig. 1). This iterative approach allows us to explore the direct feedback of human valuations on the interconnected environmental, social, and hydrological system and how perceptions of different domains may impact the choice of local 485 mitigation strategy (Khan et al., 2017).

Discussion
The case study of the Addicks and Barker reservoir system highlights the potential risk and complex interactions involving flood control dams. Here, we examined ongoing mitigation strategies associated with dam structures that had been classified as highly hazardous and had been challenged by an unprecedented amount of rainfall during Hurricane Harvey. The 90,000 490 dams within the United States have an average age of 60+ years (USACE, 2017). These aging structures, as well as the hundreds of thousands of dams throughout the world, risk structural failure and subsequent catastrophic flooding in light of increased climate intensification. We necessitate a streamlined approach to considering environmental contamination, habitat disruption, social vulnerability, and other stakeholder-driven factors when choosing how best to mitigate aging dams, especially considering the substantial costs associated with new infrastructure and cascading social/environmental effects. 495 By incorporating the inter-disciplinary science of hydrological modeling with social vulnerability and environmental risk, it is possible to consider conflicting demands and tradeoffs across the flood control domain. The CBA+MCDA method presented in this paper may be used to integrate robust stormwater modeling scenarios with multiple risk factors to better understand the correlation of hydrologic systems with overall vulnerabilities. Such comprehensive indicators of risk provide insight into the regional effects of large-scale mitigation decisions regarding extreme storm events. By linking traditional hydrological 500 modeling with GIS-based risk assessments, the differential impacts of flooding on a population can be analyzed over space for informed mitigation strategies. This approach to dam management and planning integrates the decision-maker as an explicit endogenous driver to the holistic human-water-environment system in response to increasingly complex storms, societies, and environments. When taken, such approaches will enhance our ability to form actionable insights regarding community resiliency. 505 Here we used the GIS-based SAW MCDA approach as a proof of concept to showcase how considering multiple spatial criteria within a flood management framework can improve stakeholder visualization and discussion of mitigation options that may have been discarded if viewed strictly through the lens of CBA. We encourage future research regarding additional MCDA approaches, methods for assigning stakeholder weights, and location-specific criterions to better understand the level of detail necessary for adequately framing the discussion and ensuring the integration of environmental and social domains in flood 510 risk management workflows. While the operational manuals for the studied reservoir system contained guidance for emergency-induced surcharge releases (USACE, 2012), such drastic measures had never before been necessary prior to the unprecedented rainfall observed during Hurricane Harvey. As climate change continues to stress aging dam structures, and as populations continue to densify around urban centers, we anticipate that typical operating procedures for flood control dams will become increasingly challenged. We, therefore, must consider both the soft approaches and the traditional hard-scale 515 engineering solutions for dam management, which will require an extension of the CBA paradigm to consider both the humans being impacted by the proposed alternatives and also the environments in which the systems reside. Figure A1: General timeline of Addicks and Barker Reservoir construction and major storm events, interspersed with warning reports highlighting the risks of the dams overtopping and/or necessitating emergency-induced surcharge conditions into the receiving channel (HCFCD, 1994;HCFCD, 2015;USACE, 2008). 780 Figure A2: Addicks and Barker Reservoir System (ABRS) of regional inter-connected watersheds in Houston, Texas, USA, including components of the original 1940 project plan that were later discarded (USACE, 1940). 790 Figure A4: Comparison of HEC-HMS output hydrographs at Addicks Reservoir (HEC-HMS node U1000000_9901_J) for Alternatives A1 and A2. The storage volume for A1 is 309,870.1 ac-ft, while the hydrograph volume for A2 is 258,989.6, thereby elucidating the difference in total inflow volume at Addicks Reservoir given the addition of a hypothetical third reservoir.